Skip to main content

Molecular Trajectories Before Bone Loss: A Prospective Longitudinal Protocol for Testing Whether Serial Epigenetic, Proteomic, and Microbial Measurements Improve Prediction of Peri-Implant Disease Progression

Paper type
Prospective clinical research protocol (longitudinal cohort with prediction-model development and validation)
Status
Original Work · Working Paper
Contents
  1. 1Introduction
  2. 2Background and Current Scientific Evidence
  3. 3Research Gap
  4. 4Research Objective and Central Hypothesis
  5. 5Scientific and Theoretical Framework
  6. 6Methodology
  7. 7Analytical Development
  8. 8Discussion
  9. 9Clinical Significance
  10. 10Limitations
  11. 11Future Validation
  12. 12Conclusion

Abstract

Background. Peri-implantitis is diagnosed when bone has already been lost. Molecular markers in peri-implant crevicular fluid, the submucosal microbiome, and tissue epigenetic marks differ between healthy and diseased implants in cross-sectional studies. Whether any of them changes before bone loss becomes measurable, and whether such change improves prediction beyond clinical findings, is unknown. Existing prediction models for peri-implantitis rely on clinical risk factors and have seldom been evaluated for calibration or clinical utility.

Objective. To specify a prospective cohort that tests whether within-implant trajectories of a prespecified molecular panel precede radiographic bone loss and add predictive value to a clinical model.

Design. Multicenter cohort of implants enrolled at prosthesis delivery, with crevicular fluid, submucosal plaque, and mucosal cytology sampled every six months and standardized radiographs annually for five years. The outcome is incident peri-implantitis by the 2017 World Workshop definition with a bone-loss threshold set above measurement error. A targeted panel is analyzed in the full cohort; discovery proteomics, metagenomics, and methylation profiling are performed in a nested case-control sample using specimens banked before outcome ascertainment.

Analysis. Joint models of longitudinal markers and time to event yield dynamic predictions updated at each visit. The clinical model is compared with models adding marker level and marker slope, using discrimination, calibration, and decision-curve analysis with internal–external cross-validation by center. Lead time is estimated as the interval between a marker’s departure from the individual baseline and radiographic detection.

Central hypothesis. Adding the slope of a small set of markers to clinical predictors improves calibration and net benefit for predicting peri-implantitis within the following 12 months.

Conclusions. This is a protocol. It makes no mechanistic claim and reports no data. The term “clock” is used for a composite trajectory score whose validity is the question under study.

Keywords: peri-implantitis; biomarkers; peri-implant crevicular fluid; microbiome; DNA methylation; prediction model; joint model; calibration

1. Introduction

The case definition of peri-implantitis requires bone loss. With baseline radiographs, it is bleeding or suppuration on probing, increased probing depth, and bone loss beyond the changes of initial remodeling; without them, it is a bone level at least 3 mm apical to the most coronal intraosseous part of the implant together with deep probing and bleeding [1, 2]. The diagnosis therefore confirms damage that has already happened. Radiographs add a further delay: intraoral films underestimated bone loss measured at surgery by 0.6 to 0.7 mm on average [3], and radiographic assessment of crestal bone has recognized limits of precision [4]. Since bone loss in affected implants accelerates and begins early, with more than half of eventually affected implants showing loss above 0.5 mm by the second year [5], the interval during which disease is active and undetected may be substantial.

Clinical findings before bone loss are sensitive and non-specific. Bleeding on probing is common at implants that never lose bone. Risk indicators such as a history of periodontitis, poor plaque control, and lack of maintenance identify groups at elevated risk [6, 7], and a risk assessment tool combining eight parameters has been proposed [8], but these are largely static characteristics of the patient and prosthesis and do not indicate when a particular implant is beginning to fail.

Molecular measurement is the obvious candidate to fill the gap, and a considerable literature exists. Systematic reviews find that crevicular fluid from implants with peri-implantitis contains higher levels of interleukin-1β and other proinflammatory cytokines than fluid from healthy implants [9, 10]. Panels combining markers of inflammation, matrix degradation, and bone turnover with microbial profiles distinguish disease from health [11, 12]. A chairside test for active matrix metalloproteinase-8 identified peri-implant disease with sensitivity of 71.7% and specificity of 77.8% at a threshold of 20 ng/mL in one study [13]. The peri-implantitis microbiome is depleted of commensals and enriched in pathogens [14]. Epigenetic differences have been sought [15–17].

Nearly all of this evidence compares implants that already have the disease with implants that do not. A marker that separates established peri-implantitis from health is a diagnostic marker for a condition that can be diagnosed with a probe and a radiograph. Showing that a marker changes before bone loss requires following healthy implants forward, sampling them repeatedly, and waiting. Narrative reviews note the promise of biomarkers for prognosis and the scarcity of longitudinal data [18, 19].

This protocol describes such a study. Its question is predictive and statistical: do serial molecular measurements, and in particular their rate of change within an implant, improve the prediction of incident peri-implantitis beyond what clinical examination provides? It does not ask why the markers change, and no causal interpretation is intended.

2. Background and Current Scientific Evidence

Host-derived markers in crevicular fluid

The cross-sectional evidence is extensive and of moderate quality. A meta-analysis found significant differences in interleukin-1β between healthy sites and both mucositis and peri-implantitis [9]; a systematic review found moderate evidence for higher proinflammatory cytokines at peri-implantitis [10]. Markers of bone resorption (RANK, soluble RANKL, osteoprotegerin, cathepsin K, sclerostin) were evaluated in 164 participants across health, mucositis, and peri-implantitis [12]. A 10-year follow-up measured host-derived markers at 504 implants and adjacent teeth [20]. Active matrix metalloproteinase-8 has been developed into point-of-care tests and reviewed repeatedly [13, 21–23]. Mass-spectrometric proteomics has been applied to peri-implant fluids and tissues and has helped stratify health and disease [24]. A systematic review examined marker expression before and after treatment of peri-implantitis [25].

Common weaknesses are case-control sampling from clinic populations, which inflates apparent accuracy; inconsistent normalization of crevicular fluid measures (per site, per unit time, per volume, per total protein); single time points; and no adjustment for inflammation at adjacent teeth. Markers of inflammation rise in mucositis as well as in peri-implantitis, so a single measurement cannot distinguish a stable, inflamed implant from one that is beginning to lose bone.

Microbial markers

Sequencing studies have characterized the peri-implant microbiome in health and disease: lower diversity than subgingival biofilm with distinct signatures [26]; commensal depletion and pathogen enrichment in peri-implantitis [14]; and differing co-occurrence networks of transcriptionally active taxa in peri-implantitis and periodontitis [27]. The community-level etiology has been reviewed [28]. A recent cross-sectional study related the composition and function of the submucosal microbiome to disease severity [29], and a longitudinal study described shifts during progression and recovery [30]. The longitudinal study is the nearest precedent for serial sampling. From its abstract, it followed diseased sites through treatment; it did not follow healthy implants to disease onset.

Epigenetic markers

DNA methylation in peri-implant tissues has been studied in small samples. Global methylation was examined in relation to titanium and peri-implantitis in a case-control study [15]. An exploratory study found similar global methylation between implants failing from peri-implantitis and periodontally healthy controls, with differences between gingival tissue and bone regardless of group [16]. Reviews summarize methylation and other epigenetic mechanisms in periodontal and peri-implant disease and note the possibility of predictive methylation profiles [17, 31, 32]. Methylation at defined sites tracks chronological age across tissues, the basis of epigenetic clocks [33]. Whether site-specific acceleration of such a clock occurs in chronically inflamed peri-implant mucosa has not been tested in any study identified. The single-cell characterization of rapidly advancing peri-implantitis has identified endothelial and metabolic programs that may yield further candidates [34].

Existing prediction models

Models for peri-implantitis have been built from clinical risk factors. One, in patients with periodontitis, reported an area under the curve of 0.911 in training and 0.823 in validation, with a calibration curve and decision curve [35]. A scoping review of artificial intelligence for implant prognosis found models based on records and radiographs [36]. Cross-sectional analyses of local and prosthetic factors identify predictors that could enter such models [37, 38]. These models classify existing disease or predict it from baseline characteristics. None, as far as could be determined, uses repeated molecular measurements, and few report the calibration and clinical-utility analyses that current guidance requires [39–41].

Methodological standards

Prediction research has well-defined standards: reporting by TRIPOD+AI [39]; risk-of-bias assessment by PROBAST and its extension [42, 43]; sample-size calculation that limits overfitting [44]; evaluation by discrimination, calibration, and net benefit [40, 41, 45]; and, for biomarkers, prospective collection of specimens with blinded retrospective evaluation in a nested case-control sample [46]. Joint models of longitudinal and time-to-event data allow an individual’s predicted risk to be updated as new measurements arrive [47]. These methods have been applied in other fields and are available for this one.

Summary

Cross-sectional differences in host, microbial, and epigenetic markers between healthy and diseased implants are established with moderate certainty. Longitudinal evidence that any marker precedes bone loss is lacking. Existing prediction models use static clinical variables. Methods for dynamic prediction and for rigorous evaluation exist and have not been applied to this problem.

3. Research Gap

The gap is the absence of a prospective inception cohort with serial molecular sampling of initially healthy implants, an outcome defined above measurement error, and an analysis that asks the question that matters for clinical use: does the marker add to what the clinician already knows, by how much, and at what decision threshold?

Three specific questions are unanswered. Do marker trajectories within an implant diverge before radiographic bone loss, and how long before? Is the slope of a marker more informative than its level? And does a model that includes molecular trajectories give better-calibrated risks and greater net benefit than a clinical model?

4. Research Objective and Central Hypothesis

The primary objective is to develop and internally validate a dynamic prediction model for incident peri-implantitis that incorporates serial molecular measurements, and to quantify its incremental value over a clinical model.

The central hypothesis is that a model including the current level and the within-implant slope of a prespecified targeted panel, in addition to clinical predictors, predicts peri-implantitis within the next 12 months with better calibration and higher net benefit across a clinically relevant range of threshold probabilities than the clinical model alone.

Secondary hypotheses: (a) for at least one marker, the within-implant slope is associated with the hazard of peri-implantitis independently of the current level; (b) among implants that develop peri-implantitis, the composite trajectory score departs from the individual baseline at least one sampling interval before the radiographic criterion is met; (c) a mucosal epigenetic age-acceleration measure is higher at implants that later develop disease (exploratory).

The central hypothesis is refuted if the extended model shows no improvement in calibration or net benefit on cross-validation. That outcome would be informative: it would indicate that the cost and complexity of serial molecular sampling are not justified for prediction, whatever the markers’ diagnostic associations.

5. Scientific and Theoretical Framework

Why trajectories and not levels

Crevicular fluid markers vary widely between individuals at health. For an analyte with large between-subject and smaller within-subject variation, a value that is unremarkable against a population reference range can represent a large departure from that individual’s own baseline. Laboratory medicine expresses this through the index of individuality, the ratio of within-subject to between-subject variation; when it is low, population-based cutoffs are insensitive and subject-based reference changes are more informative. Peri-implant markers have not been characterized in these terms. If their individuality is low, as the wide cross-sectional overlap between groups suggests, the informative signal is change relative to self. That requires a baseline at health and repeated sampling, which a cross-sectional design cannot supply.

The composite trajectory score

The term “molecular clock” in the working title of this project refers to a composite score that summarizes how far, and how fast, an implant’s molecular profile has moved from its own healthy baseline. For marker j measured at visit time t in implant i, a linear mixed model on the log scale gives

logy_(ij)(t)=β_(0j)+b_(0ij)+(β_(1j)+b_(1ij)) t+x_i(t)^⊤γ_j+ε_(ij)(t)  (1)

with fixed intercept and slope β_(0j),β_(1j), implant-specific random intercept and slope b_(0ij),b_(1ij), time-varying covariates x_i(t) (sampling conditions, plaque at the visit, recent professional cleaning), and residual ε_(ij). The model-based current value m_(ij)(t) and slope m′_(ij)(t) are the inputs to prediction. The composite score is a weighted sum of standardized slopes across markers, with weights estimated in the survival submodel. The clock metaphor is apt only if the score advances monotonically toward the event, which is one of the things the study tests. No biological mechanism is implied.

Dynamic prediction

The hazard of peri-implantitis for implant i at time t is modeled as

h_i(t)=h_0(t) exp​(z_i^⊤η+∑_j^​ α_(1j) m_(ij)(t)+∑_j^​ α_(2j) m′_(ij)(t))  (2)

with baseline hazard h_0, baseline clinical predictors z_i, and association parameters α_(1j) for current value and α_(2j) for slope. Fitting equations (1) and (2) jointly accounts for measurement error in the markers and for informative dropout at the event [47]. For an implant event-free at visit time t with marker history Y_i(t), the model yields the conditional probability of an event before t+Δ:

π_i(t+Δ∣t)=Pr(T_i≤t+Δ∣T_i>t, Y_i(t), z_i)  (3)

This is the quantity a clinician would use at a recall visit, with Δ of 12 months as the primary horizon.

6. Methodology

Design and setting

A prospective, multicenter inception cohort in university and private periodontal and prosthodontic practices, reported to STROBE [48] and TRIPOD+AI [39]. The protocol is registered before enrollment, and the statistical analysis plan is finalized before any outcome data are examined.

Participants

Inclusion. Adults receiving one or more bone-level or tissue-level titanium implants restored with fixed prostheses; enrollment at delivery of the definitive prosthesis; peri-implant health at enrollment by current criteria [1]; willingness to attend six-monthly visits for five years.

Exclusion. Untreated periodontitis at enrollment; systemic antibiotic or anti-inflammatory therapy within three months of a sampling visit (the visit is deferred, not the participant excluded); conditions precluding standardized radiography.

Sampling of implants. One index implant per participant for the primary analysis, selected at random when more than one is eligible, to avoid clustering in the primary model; additional implants are followed for secondary analyses with patient-level random effects.

Predictors

Clinical. History of periodontitis and its stage and grade; smoking; diabetes and HbA1c; plaque and bleeding indices; probing depths; keratinized mucosa width; implant system, surface, and connection; prosthesis type, retention, and emergence profile; access for cleaning; adherence to maintenance. These are the variables in existing risk tools and models [8, 35, 37].

Targeted molecular panel (full cohort). Selected before enrollment from markers with the most consistent cross-sectional evidence and available validated assays: interleukin-1β, active matrix metalloproteinase-8, soluble RANKL and osteoprotegerin, and crevicular fluid volume [9, 10, 12, 21]; and a small set of microbial indicators quantified by targeted molecular assay, chosen from taxa consistently enriched or depleted in peri-implantitis [14, 26]. The panel is deliberately small, to limit the number of parameters.

Discovery assays (nested sample). Mass-spectrometric proteomics of crevicular fluid [24]; shotgun metagenomics or full-length amplicon sequencing of submucosal plaque with standard processing [49]; and DNA methylation profiling of mucosal cells collected by brush cytology, with an oral-epithelial reference sample from the buccal mucosa of the same participant to allow within-person comparison and computation of epigenetic age acceleration [33].

Specimen collection and handling

Sampling follows a written standard operating procedure. Visits are scheduled at a consistent time of day; participants refrain from eating, drinking, and oral hygiene for a fixed period beforehand. The site is isolated and supragingival plaque removed. Crevicular fluid is collected on paper strips for a fixed time at fixed sites before probing; volume is measured electronically; strips visibly contaminated with blood are recorded and replaced. Submucosal plaque is collected with sterile paper points after fluid sampling. Mucosal cytology follows. Specimens are aliquoted and frozen at −80 °C within a fixed interval. Pre-analytical variables (time to freezing, operator, strip lot) are recorded. A duplicate sample from a second site at each visit in a random subset estimates within-visit variability.

Outcome

The primary outcome is time to incident peri-implantitis at the index implant, defined as bleeding or suppuration on probing with radiographic bone loss of at least 1.0 mm relative to the radiograph taken one year after prosthesis delivery, the latter serving as the post-remodeling baseline [1]. The 1.0-mm threshold is set above reported radiographic measurement error and underestimation [3, 4] and is examined in sensitivity analyses at 0.5 mm and 2.0 mm. Radiographs are taken with individualized film holders and read by two calibrated examiners blinded to clinical and molecular data; disagreements are adjudicated. An event date is assigned to the visit at which the criterion is first met. Implant loss from peri-implantitis counts as an event; loss from other causes is a competing event.

Sample size

Sample size is set by the requirements for developing a prediction model without excessive overfitting [44]. For a time-to-event model, the number of participants needed to target a global shrinkage factor S with p candidate predictor parameters and an anticipated Cox–Snell R_CS^2 is

n=p/((S−1) ln​(1−R_CS^2/S))  (4)

With p=10 parameters for the extended model, S=0.9, and a conservative R_CS^2 of 0.05, equation (4) gives approximately 1,750 participants. The value of R_CS^2 is an assumption. If the five-year cumulative incidence at the index implant is on the order of 10%, also an assumption to be checked against accruing data, about 175 events would be observed, or 17 events per parameter. Allowing 20% loss to follow-up, enrollment of about 2,200 is planned. Incidence will be monitored without unblinding predictors, and follow-up extended if events accrue more slowly than anticipated.

The nested discovery sample comprises all incident cases and two controls per case matched on center, time in function, and history of periodontitis, selected by incidence-density sampling.

Statistical analysis

Models. Three nested models are developed. Model C contains clinical predictors only, with time-varying clinical indices handled by the same joint framework. Model L adds the current level of each targeted marker. Model S adds the slope. Functional forms are prespecified; continuous predictors are not dichotomized [50].

Performance. For predictions made at each landmark visit for the following 12 months: time-dependent area under the curve and concordance; calibration intercept, slope, and flexible calibration curves [40]; Brier score; and net benefit across threshold probabilities from 5% to 30%, a range within which intensified maintenance or early intervention might reasonably be triggered [41]. Net benefit at threshold p_t is

NB(p_t)=TP/n−FP/n⋅p_t/(1−p_t)  (5)

with TP and FP the numbers of true and false positives at that threshold among n at risk.

Validation. Internal validation by bootstrap with optimism correction, and internal–external cross-validation in which each center is held out in turn. Heterogeneity of performance across centers is reported.

Incremental value. The primary comparison is Model S against Model C on calibration and net benefit. Differences in concordance are reported with confidence intervals and not used as the sole criterion, since concordance is insensitive to improvements in calibration.

Lead time. Among cases, the visit at which the composite score first exceeds the implant’s own baseline by a prespecified reference change is identified, and the interval to the event visit is summarized. The same rule is applied to controls to estimate the false-alarm rate per implant-year. This analysis is descriptive.

Discovery analyses. In the nested sample, markers are evaluated by conditional models with false discovery rate control; candidates are carried forward only if they can be measured by a targeted assay and are then tested in the remainder of the cohort, which serves as an independent validation set for those markers [46].

Missing data. Missed visits are handled by the joint model under a missing-at-random assumption; missing baseline covariates are multiply imputed [51]. Sensitivity analyses examine departures from these assumptions.

Confounding, bias, and blinding

Because the aim is prediction, confounding in the causal sense is not the concern; the concern is bias in estimated performance. PROBAST domains are addressed as follows [42, 43]. Participants: an inception cohort avoids the spectrum bias of case-control sampling. Predictors: assays are run blinded to outcome, in batches balanced for center and visit, with bridging samples to control drift. Outcome: adjudicated blind to molecular data, with a fixed definition. Analysis: adequate events per parameter, prespecified models, no data-driven cutpoints, and appropriate handling of missing data and competing events. Inflammation at adjacent teeth, recent professional cleaning, and plaque at the visit affect markers and are recorded as covariates in equation (1).

Ethics and data management

Institutional review board approval at each center and written consent, including consent for genomic and epigenomic analysis of human and microbial material and for controlled-access data sharing. Participants receive standard maintenance throughout; the study does not withhold care. Any implant meeting clinical criteria for disease is treated according to current guidelines [52]. Molecular results are not returned to clinicians during the study, because their meaning is unknown and returning them would alter care in ways that bias the outcome.

7. Analytical Development

How much lead time is available in principle

Let radiographic detection require bone loss of at least θ beyond baseline, and let bone be lost at rate ρ(t) after biological onset at time t_0. Detection occurs at t_d satisfying ∫_t_0^t_d ρ(t) dt=θ. A marker that responds promptly to the onset of resorption has a potential lead time L=t_d−t_0. For a constant early rate ρ,

L=θ/ρ  (6)

With θ=1.0 mm and an early rate of 0.5 mm per year, L is two years; with a rate of 0.25 mm per year, four years. These figures are illustrations of equation (6), not estimates. Reported bone-loss patterns are non-linear and accelerating [5], which implies low early rates and, by this reasoning, a long window in which a responsive marker could lead radiographic detection. The window exists only if markers respond to early resorption and are not merely reporting mucosal inflammation. Six-monthly sampling resolves lead times of a year or more; shorter lead times would go undetected by this design.

How many visits are needed to estimate a slope

For a marker with within-implant standard deviation σ_w on the log scale (biological plus analytical), measured at k equally spaced visits separated by interval τ, the standard error of the estimated slope is

SE(β)=σ_w/τ√(12/(k (k^2−1)))  (7)

The standard error falls roughly as k^(−3/2). With three visits, it is 0.71 σ_w/τ; with five, 0.32 σ_w/τ. A true slope β is distinguishable from zero at about two standard errors, so a marker that rises by one within-implant standard deviation per year (β=σ_w per year, with τ=0.5 year) needs about five visits, or two years of history, before its slope is reliably non-zero. Slopes therefore become informative only after an implant has been in the study for some time, and predictions in the first year rest on levels. The joint model handles this automatically by shrinking uncertain slopes toward the population mean. The design implication is that within-implant variability must be measured early, in the duplicate-sample substudy, since the feasibility of slope-based prediction depends on it.

Reference change for an individual implant

For two measurements on the same implant, the difference that exceeds analytical and within-subject biological variation with probability 1−α is the reference change value. On the log scale,

RCV=z_(1−α/2) √(2) σ_w  (8)

A lead-time rule based on departure from individual baseline is defined in these units. If σ_w for a marker is large relative to the difference between health and early disease, no rule based on that marker can have both a useful lead time and a tolerable false-alarm rate, regardless of how strongly the marker separates established disease from health in cross-section. This is the quantitative form of the argument that diagnostic accuracy does not imply predictive value.

What improvement would matter

A model can improve discrimination trivially and remain clinically useless. The decision at stake at a recall visit is whether to intensify maintenance or investigate further. If intensified maintenance is low-cost and low-harm, the relevant threshold probability is low, perhaps 5%–10%, and at such thresholds a treat-all strategy is hard to beat. The extended model is worth its cost only if its net benefit exceeds that of the clinical model, and of treat-all, by a margin that offsets the burden of sampling and assay. That margin can be expressed as the number of unnecessary intensifications avoided per 100 implants at equal detection, which follows from the difference in net benefit divided by p_t/(1−p_t). The analysis plan prespecifies reporting in these terms.

8. Discussion

The protocol addresses a question that the biomarker literature has approached without answering. The cross-sectional evidence shows that diseased and healthy implants differ at the molecular level. It does not show that molecular measurement would help a clinician who already has a probe, a radiograph, and a patient history. Establishing that requires an inception cohort, repeated sampling, an outcome above measurement error, and an analysis of incremental value, which is what is proposed.

There are good reasons to expect a modest result. Inflammatory markers in crevicular fluid track mucosal inflammation, which bleeding on probing already measures cheaply; their marginal information may be small. Within-implant variability of crevicular fluid analytes is likely to be high, given sensitivity to sampling technique, plaque at the visit, and time since cleaning, and equations (7) and (8) show how quickly that erodes the value of a slope. Bone-turnover markers are more specific to the event of interest and present at lower concentrations with greater assay noise. Microbial markers vary between individuals in ways that may swamp within-implant change. A finding that the extended model adds little would be consistent with experience in other fields, where biomarkers with strong case-control associations have often failed to improve prediction.

There are also reasons to think trajectories could succeed where levels have not. Using each implant as its own control removes between-person variation, the dominant source of overlap in cross-sectional data. A slope integrates over several visits. And the accelerating course of bone loss implies an early phase of slow, undetected resorption during which a responsive marker would have time to register.

The choice to prespecify a small targeted panel and confine high-dimensional assays to a nested discovery sample is deliberate. Developing a model from thousands of omic features in a cohort with fewer than 200 events would produce optimistic and irreproducible results regardless of the algorithm. The two-phase design allows discovery while protecting the primary analysis, and uses the remainder of the cohort to validate anything discovered.

The epigenetic component is the most speculative. The original framing of this project invoked a molecular clock, and the nearest rigorous counterpart is epigenetic age acceleration measured in mucosal cells at the implant relative to a reference site in the same mouth. The existing methylation studies in peri-implant disease are small and inconsistent [15, 16], and cell-type composition confounds methylation in mixed samples. The analysis is exploratory and is labeled as such.

Several design decisions trade rigor against feasibility. One index implant per participant simplifies the primary model and discards information. Six-monthly sampling matches maintenance schedules and cannot resolve short lead times. A five-year horizon is long for a study and short relative to the natural history of the disease. Withholding molecular results from clinicians is necessary for validity and means the study cannot show that acting on the predictions improves outcomes; that would require a subsequent randomized trial of prediction-guided care.

The study estimates predictive associations. A marker that predicts is not thereby shown to participate in the disease process, and a marker that fails to predict is not shown to be irrelevant to it. Questions of mechanism, and of whether transitions between reversible and persistent disease have a threshold character, are separate and need different designs.

9. Clinical Significance

If the central hypothesis is supported, the cohort will deliver a validated dynamic model, an estimate of how much earlier than radiography it identifies progressing implants, and the false-alarm rate at which it does so. Those three quantities determine whether serial molecular monitoring is worth offering, to whom, and at what interval. A chairside assay for one or two markers whose slope proved informative would be the practical endpoint.

If the hypothesis is not supported, the result will redirect effort. Resources now spent validating diagnostic markers against established disease could move to improving the clinical model, to standardizing radiographic surveillance, or to other measurement modalities.

In either case the study provides data the field lacks: within-implant biological variation of commonly measured analytes, the incidence and timing of peri-implantitis from an inception cohort with standardized ascertainment, and a rigorously evaluated clinical prediction model.

10. Limitations

The cohort is large, long, and expensive, and sample-size assumptions about incidence and model fit may be wrong. Attrition over five years could be substantial and informative. The outcome threshold of 1.0 mm trades sensitivity for specificity; early cases will be dated later than their true onset, which inflates apparent lead time, and the sensitivity analyses address this only partly. Crevicular fluid sampling is technique-sensitive and may not standardize across centers. The targeted panel reflects current evidence and may omit the markers that matter. Brush cytology yields mixed cell populations and limited DNA. Participants in a study with six-monthly professional attention may have lower incidence than the general population, reducing events and limiting generalizability. External validation in an independent cohort is beyond the scope of the protocol. The literature review was conducted from abstracts and index records.

11. Future Validation

External validation of the final model in a separately recruited cohort, with different centers and assay laboratories, is required before any clinical use. If performance holds, an impact study would randomize clinics or patients to prediction-guided recall versus standard recall, with peri-implantitis incidence and resource use as outcomes. Any marker emerging from the discovery phase needs analytical validation of a targeted assay and confirmation in the validation portion of the cohort. Data and code should be deposited to allow reanalysis and updating as methods evolve.

12. Conclusion

Molecular differences between healthy and diseased implants are well documented, and their value for anticipating disease is not. This protocol specifies a prospective cohort that follows healthy implants with serial sampling, defines the outcome above measurement error, models marker trajectories jointly with time to disease, and evaluates whether they improve calibrated, clinically useful prediction beyond clinical findings. Simple analytical relations show that the answer depends on within-implant variability and on the early rate of bone loss, neither of which has been measured. The study would measure both. It reports no results and asserts no mechanism.

Cite this paper

Jain S, Gupta A. Molecular Trajectories Before Bone Loss: A Prospective Longitudinal Protocol for Testing Whether Serial Epigenetic, Proteomic, and Microbial Measurements Improve Prediction of Peri-Implant Disease Progression. Working paper. Center for Implant Dentistry; 2026. Available from: https://bayareaimplantdentistry.com/clinician-research-and-publications/molecular-clock-peri-implant-disease

References

  1. Renvert S, Persson GR, Pirih FQ, Camargo PM. Peri-implant health, peri-implant mucositis, and peri-implantitis: Case definitions and diagnostic considerations. J Periodontol. 2018;89 Suppl 1:S304-S312. doi:10.1002/JPER.17-0588 (opens in a new tab). PMID: 29926953 (opens in a new tab).
  2. Berglundh T, Armitage G, Araujo MG, Avila-Ortiz G, Blanco J, Camargo PM, et al. Peri-implant diseases and conditions: Consensus report of workgroup 4 of the 2017 World Workshop on the Classification of Periodontal and Peri-Implant Diseases and Conditions. J Clin Periodontol. 2018;45 Suppl 20:S286-S291. doi:10.1111/jcpe.12957 (opens in a new tab). PMID: 29926491 (opens in a new tab).
  3. Serino G, Sato H, Holmes P, Turri A. Intra-surgical vs. radiographic bone level assessments in measuring peri-implant bone loss. Clin Oral Implants Res. 2017;28(11):1396-1400. doi:10.1111/clr.12999 (opens in a new tab). PMID: 28009061 (opens in a new tab).
  4. De Bruyn H, Vandeweghe S, Ruyffelaert C, Cosyn J, Sennerby L. Radiographic evaluation of modern oral implants with emphasis on crestal bone level and relevance to peri-implant health. Periodontol 2000. 2013;62(1):256-70. doi:10.1111/prd.12004 (opens in a new tab). PMID: 23574471 (opens in a new tab).
  5. Derks J, Schaller D, Håkansson J, Wennström JL, Tomasi C, Berglundh T. Peri-implantitis - onset and pattern of progression. J Clin Periodontol. 2016;43(4):383-8. doi:10.1111/jcpe.12535 (opens in a new tab). PMID: 26900869 (opens in a new tab).
  6. Schwarz F, Derks J, Monje A, Wang HL. Peri-implantitis. J Periodontol. 2018;89 Suppl 1:S267-S290. doi:10.1002/JPER.16-0350 (opens in a new tab). PMID: 29926957 (opens in a new tab).
  7. Costa FO, Takenaka-Martinez S, Cota LO, Ferreira SD, Silva GL, Costa JE. Peri-implant disease in subjects with and without preventive maintenance: a 5-year follow-up. J Clin Periodontol. 2012;39(2):173-81. doi:10.1111/j.1600-051X.2011.01819.x (opens in a new tab). PMID: 22111654 (opens in a new tab).
  8. Heitz-Mayfield LJA, Heitz F, Lang NP. Implant Disease Risk Assessment IDRA-a tool for preventing peri-implant disease. Clin Oral Implants Res. 2020;31(4):397-403. doi:10.1111/clr.13585 (opens in a new tab). PMID: 32003037 (opens in a new tab).
  9. Faot F, Nascimento GG, Bielemann AM, Campão TD, Leite FR, Quirynen M. Can peri-implant crevicular fluid assist in the diagnosis of peri-implantitis? A systematic review and meta-analysis. J Periodontol. 2015;86(5):631-45. doi:10.1902/jop.2015.140603 (opens in a new tab). PMID: 25675962 (opens in a new tab).
  10. Duarte PM, Serrão CR, Miranda TS, Zanatta LC, Bastos MF, Faveri M, et al. Could cytokine levels in the peri-implant crevicular fluid be used to distinguish between healthy implants and implants with peri-implantitis? A systematic review. J Periodontal Res. 2016;51(6):689-698. doi:10.1111/jre.12354 (opens in a new tab). PMID: 26774043 (opens in a new tab).
  11. Wang HL, Garaicoa-Pazmino C, Collins A, Ong HS, Chudri R, Giannobile WV. Protein biomarkers and microbial profiles in peri-implantitis. Clin Oral Implants Res. 2016;27(9):1129-36. doi:10.1111/clr.12708 (opens in a new tab). PMID: 26424287 (opens in a new tab).
  12. Rakic M, Struillou X, Petkovic-Curcin A, Matic S, Canullo L, Sanz M, et al. Estimation of bone loss biomarkers as a diagnostic tool for peri-implantitis. J Periodontol. 2014;85(11):1566-74. doi:10.1902/jop.2014.140069 (opens in a new tab). PMID: 24878328 (opens in a new tab).
  13. Xanthopoulou V, Räisänen IT, Sorsa T, Tortopidis D, Sakellari D. Diagnostic value of aMMP-8 and azurocidin in peri-implant sulcular fluid as biomarkers of peri-implant health or disease. Clin Exp Dent Res. 2024;10(3):e883. doi:10.1002/cre2.883 (opens in a new tab). PMID: 38853451 (opens in a new tab).
  14. Sanz-Martin I, Doolittle-Hall J, Teles RP, Patel M, Belibasakis GN, Hämmerle CHF, et al. Exploring the microbiome of healthy and diseased peri-implant sites using Illumina sequencing. J Clin Periodontol. 2017;44(12):1274-1284. doi:10.1111/jcpe.12788 (opens in a new tab). PMID: 28766745 (opens in a new tab).
  15. Daubert DM, Pozhitkov AE, Safioti LM, Kotsakis GA. Association of Global DNA Methylation to Titanium and Peri-Implantitis: A Case-Control Study. JDR Clin Trans Res. 2019;4(3):284-291. doi:10.1177/2380084418822831 (opens in a new tab). PMID: 30931716 (opens in a new tab).
  16. Khouly I, Pardiñas López S, Díaz Prado SM, Ferrantino L, Kalm J, Larsson L, et al. Global DNA Methylation in Dental Implant Failure Due to Peri-Implantitis: An Exploratory Clinical Pilot Study. Int J Environ Res Public Health. 2022;19(2). doi:10.3390/ijerph19021020 (opens in a new tab). PMID: 35055840 (opens in a new tab).
  17. Larsson L, Giraldo-Osorno PM, Garaicoa-Pazmino C, Giannobile WV, Asa’ad F. DNA and RNA Methylation in Periodontal and Peri-implant Diseases. J Dent Res. 2025;104(2):131-139. doi:10.1177/00220345241291533 (opens in a new tab). PMID: 39629934 (opens in a new tab).
  18. Alassy H, Parachuru P, Wolff L. Peri-Implantitis Diagnosis and Prognosis Using Biomarkers in Peri-Implant Crevicular Fluid: A Narrative Review. Diagnostics (Basel). 2019;9(4). doi:10.3390/diagnostics9040214 (opens in a new tab). PMID: 31817894 (opens in a new tab).
  19. Kinane DF, Bıyıkoğlu B. Diagnostic Biomarkers of Periodontitis and Peri-Implantitis: The Past, the Present, the Future. J Periodontal Res. 2026;61(9):997-1008. doi:10.1111/jre.70140 (opens in a new tab). PMID: 42473265 (opens in a new tab).
  20. Ramseier CA, Eick S, Brönnimann C, Buser D, Brägger U, Salvi GE. Host-derived biomarkers at teeth and implants in partially edentulous patients. A 10-year retrospective study. Clin Oral Implants Res. 2016;27(2):211-7. doi:10.1111/clr.12566 (opens in a new tab). PMID: 25682848 (opens in a new tab).
  21. Lähteenmäki H, Pätilä T, Pärnänen CP, Räisänen I, Tervahartiala T, Gupta S, et al. aMMP-8 point-of-care - diagnostic methods and treatment modalities in periodontitis and peri-implantitis. Expert Opin Ther Targets. 2023;27(7):627-637. doi:10.1080/14728222.2023.2240014 (opens in a new tab). PMID: 37522314 (opens in a new tab).
  22. Räisänen IT, Aji NRAS, Sakellari D, Grigoriadis A, Rantala I, Pätilä T, et al. Active Matrix Metalloproteinase-8 (aMMP-8) Versus Total MMP-8 in Periodontal and Peri-Implant Disease Point-of-Care Diagnostics. Biomedicines. 2023;11(11). doi:10.3390/biomedicines11112885 (opens in a new tab). PMID: 38001886 (opens in a new tab).
  23. Alassiri S, Parnanen P, Rathnayake N, Johannsen G, Heikkinen AM, Lazzara R, et al. The Ability of Quantitative, Specific, and Sensitive Point-of-Care/Chair-Side Oral Fluid Immunotests for aMMP-8 to Detect Periodontal and Peri-Implant Diseases. Dis Markers. 2018;2018:1306396. doi:10.1155/2018/1306396 (opens in a new tab). PMID: 30154936 (opens in a new tab).
  24. Halstenbach T, Topitsch A, Schilling O, Iglhaut G, Nelson K, Fretwurst T. Mass spectrometry-based proteomic applications in dental implants research. Proteomics Clin Appl. 2024;18(3):e2300019. doi:10.1002/prca.202300019 (opens in a new tab). PMID: 38342588 (opens in a new tab).
  25. Moaven H, Giacaman A, Beltrán V, Sam YH, Betancur D, Mainas G, et al. Biomarker Expression of Peri-Implantitis Lesions before and after Treatment: A Systematic Review. Int J Environ Res Public Health. 2022;19(21). doi:10.3390/ijerph192114085 (opens in a new tab). PMID: 36360962 (opens in a new tab).
  26. Kumar PS, Mason MR, Brooker MR, O’Brien K. Pyrosequencing reveals unique microbial signatures associated with healthy and failing dental implants. J Clin Periodontol. 2012;39(5):425-33. doi:10.1111/j.1600-051X.2012.01856.x (opens in a new tab). PMID: 22417294 (opens in a new tab).
  27. Shiba T, Watanabe T, Kachi H, Koyanagi T, Maruyama N, Murase K, et al. Distinct interacting core taxa in co-occurrence networks enable discrimination of polymicrobial oral diseases with similar symptoms. Sci Rep. 2016;6:30997. doi:10.1038/srep30997 (opens in a new tab). PMID: 27499042 (opens in a new tab).
  28. Belibasakis GN, Manoil D. Microbial Community-Driven Etiopathogenesis of Peri-Implantitis. J Dent Res. 2021;100(1):21-28. doi:10.1177/0022034520949851 (opens in a new tab). PMID: 32783779 (opens in a new tab).
  29. Joshi AA, Szafrański SP, Steglich M, Yang I, Behrens W, Schaefer-Dreyer P, et al. The Submucosal Microbiome Correlates with Peri-implantitis Severity. J Dent Res. 2026;105(3):313-322. doi:10.1177/00220345251352809 (opens in a new tab). PMID: 40719760 (opens in a new tab).
  30. Anuntakarun S, Thaweesapphithak S, Krasaesin A, Prommanee S, Arunyanak S, Kungsadalpipob K, et al. Microbiome Shifts in Peri-Implantitis: Longitudinal Characterization of Dysbiosis and Resolution. Int Dent J. 2025;75(5):100951. doi:10.1016/j.identj.2025.100951 (opens in a new tab). PMID: 40811957 (opens in a new tab).
  31. Larsson L, Kavanagh NM, Nguyen TVN, Castilho RM, Berglundh T, Giannobile WV. Influence of epigenetics on periodontitis and peri-implantitis pathogenesis. Periodontol 2000. 2022;90(1):125-137. doi:10.1111/prd.12453 (opens in a new tab). PMID: 35913702 (opens in a new tab).
  32. Li Q, Lu H, Zhang M, Ye Y, Chen Q, Sun P. Epigenetic factors associated with peri-implantitis: a review. J Zhejiang Univ Sci B. 2025;26(7):657-674. doi:10.1631/jzus.B2400032 (opens in a new tab). PMID: 40722244 (opens in a new tab).
  33. Horvath S. DNA methylation age of human tissues and cell types. Genome Biol. 2013;14(10):R115. doi:10.1186/gb-2013-14-10-r115 (opens in a new tab). PMID: 24138928 (opens in a new tab).
  34. Easter QT, Huynh KLA, Stolf CS, Xie J, Matuck BF, Hasuike A, et al. CD38⁺ endothelial remodeling marks spatially patterned vasculopathy in rapidly advancing periodontitis and peri-implantitis. Nat Commun. 2026;17(1). doi:10.1038/s41467-026-72452-2 (opens in a new tab). PMID: 42103720 (opens in a new tab).
  35. Gao X, Li D, Duan Y, Wu L. Risk factors and prediction model of peri-implantitis in post operative periodontitis patients. Am J Transl Res. 2024;16(9):4741-4750. doi:10.62347/THTL1156 (opens in a new tab). PMID: 39398596 (opens in a new tab).
  36. Wu Z, Yu X, Wang F, Xu C. Application of artificial intelligence in dental implant prognosis: A scoping review. J Dent. 2024;144:104924. doi:10.1016/j.jdent.2024.104924 (opens in a new tab). PMID: 38467177 (opens in a new tab).
  37. Monje A, Pons R, Soldini MC, Rosen PS, Nart J, Buser D. Role of Local Factors on the Occurrence of Peri-Implantitis: A Cross-Sectional Analysis. Clin Oral Implants Res. 2025;36(12):1614-1625. doi:10.1111/clr.70036 (opens in a new tab). PMID: 40916656 (opens in a new tab).
  38. Yi Y, Koo KT, Schwarz F, Ben Amara H, Heo SJ. Association of prosthetic features and peri-implantitis: A cross-sectional study. J Clin Periodontol. 2020;47(3):392-403. doi:10.1111/jcpe.13251 (opens in a new tab). PMID: 31912511 (opens in a new tab).
  39. Collins GS, Moons KGM, Dhiman P, Riley RD, Beam AL, Van Calster B, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378. doi:10.1136/bmj-2023-078378 (opens in a new tab). PMID: 38626948 (opens in a new tab).
  40. Van Calster B, McLernon DJ, van Smeden M, Wynants L, Steyerberg EW. Calibration: the Achilles heel of predictive analytics. BMC Med. 2019;17(1):230. doi:10.1186/s12916-019-1466-7 (opens in a new tab). PMID: 31842878 (opens in a new tab).
  41. Vickers AJ, Elkin EB. Decision curve analysis: a novel method for evaluating prediction models. Med Decis Making. 2006;26(6):565-74. doi:10.1177/0272989X06295361 (opens in a new tab). PMID: 17099194 (opens in a new tab).
  42. Wolff RF, Moons KGM, Riley RD, Whiting PF, Westwood M, Collins GS, et al. PROBAST: A Tool to Assess the Risk of Bias and Applicability of Prediction Model Studies. Ann Intern Med. 2019;170(1):51-58. doi:10.7326/M18-1376 (opens in a new tab). PMID: 30596875 (opens in a new tab).
  43. Moons KGM, Damen JAA, Kaul T, Hooft L, Andaur Navarro C, Dhiman P, et al. PROBAST+AI: an updated quality, risk of bias, and applicability assessment tool for prediction models using regression or artificial intelligence methods. BMJ. 2025;388:e082505. doi:10.1136/bmj-2024-082505 (opens in a new tab). PMID: 40127903 (opens in a new tab).
  44. Riley RD, Ensor J, Snell KIE, Harrell FE Jr, Martin GP, Reitsma JB, et al. Calculating the sample size required for developing a clinical prediction model. BMJ. 2020;368:m441. doi:10.1136/bmj.m441 (opens in a new tab). PMID: 32188600 (opens in a new tab).
  45. Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N, et al. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology. 2010;21(1):128-38. doi:10.1097/EDE.0b013e3181c30fb2 (opens in a new tab). PMID: 20010215 (opens in a new tab).
  46. Pepe MS, Feng Z, Janes H, Bossuyt PM, Potter JD. Pivotal evaluation of the accuracy of a biomarker used for classification or prediction: standards for study design. J Natl Cancer Inst. 2008;100(20):1432-8. doi:10.1093/jnci/djn326 (opens in a new tab). PMID: 18840817 (opens in a new tab).
  47. Rizopoulos D. Dynamic predictions and prospective accuracy in joint models for longitudinal and time-to-event data. Biometrics. 2011;67(3):819-29. doi:10.1111/j.1541-0420.2010.01546.x (opens in a new tab). PMID: 21306352 (opens in a new tab).
  48. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet. 2007;370(9596):1453-7. doi:10.1016/S0140-6736(07)61602-X (opens in a new tab). PMID: 18064739 (opens in a new tab).
  49. Callahan BJ, McMurdie PJ, Rosen MJ, Han AW, Johnson AJ, Holmes SP. DADA2: High-resolution sample inference from Illumina amplicon data. Nat Methods. 2016;13(7):581-3. doi:10.1038/nmeth.3869 (opens in a new tab). PMID: 27214047 (opens in a new tab).
  50. Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15(4):361-87. doi:10.1002/(SICI)1097-0258(19960229)15:4<361::AID-SIM168>3.0.CO;2-4 (opens in a new tab). PMID: 8668867 (opens in a new tab).
  51. Sterne JA, White IR, Carlin JB, Spratt M, Royston P, Kenward MG, et al. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ. 2009;338:b2393. doi:10.1136/bmj.b2393 (opens in a new tab). PMID: 19564179 (opens in a new tab).
  52. Herrera D, Berglundh T, Schwarz F, Chapple I, Jepsen S, Sculean A, et al. Prevention and treatment of peri-implant diseases-The EFP S3 level clinical practice guideline. J Clin Periodontol. 2023;50 Suppl 26:4-76. doi:10.1111/jcpe.13823 (opens in a new tab). PMID: 37271498 (opens in a new tab).